systemic risk and the solvency-liquidity nexus of banks · 2017. 5. 2. · systemic risk and the...
TRANSCRIPT
Systemic risk and the solvency-liquidity nexus of banks
Diane Pierret1,2
This version: March 28, 2014
Abstract: This paper shows the empirical interaction between solvency and liquidityrisks of banks that make them particularly vulnerable to an aggregate crisis. I find thatbanks lose their access to short-term funding when markets expect they will be insolvent ina crisis. Conversely, a bank with more short-term debt (with a large exposure to fundingliquidity risk) gets a larger capital shortfall estimate. Importantly, the short-term debt ofa bank is not sensitive to the risk of the bank failing in isolation but is influenced by itssolvency risk when the whole economy is under stress (measured as the expected capitalshortfall in a crisis). This solvency-liquidity nexus is found to be strong under many robust-ness checks and to contain useful information for forecasting the short-term balance sheet ofbanks. The results suggest that the solvency-liquidity interaction should be accounted forwhen designing liquidity and capital requirements, in contrast to Basel III regulation wheresolvency and liquidity risks are treated separately.
Keywords: capital shortfall, funding liquidity risk, short-term funding.
JEL Classification: G01, G21, G28.
1NYU Stern School of Business, Volatility Institute, 44 West 4th Street, New York, NY 10012.2Université catholique de Louvain, ISBA, 20 Voie du Roman Pays, B-1348 Louvain-La-Neuve, Belgium.
E-mail: [email protected]
I am extremely grateful to Viral Acharya, Luc Bauwens, Robert Engle and Christian Hafner for theirexcellent guidance and continuous support. I thank Stephen Figlewski, Andres Liberman, Matteo Luciani,Matthew Richardson, Sascha Steffen and David Veredas for helpful comments. I also thank Rob Capellinifor providing me with V-Lab’s measures of systemic risk. All remaining errors are my own.
“A more interesting approach would be to tie liquidity and capital standards together byrequiring higher levels of capital for large firms unless their liquidity position is substantiallystronger than minimum requirements. This approach would reflect the fact that the marketperception of a given firm’s position as counterparty depends upon the combination of itsfunding position and capital levels. [...] While there is decidedly a need for solid minimumrequirements for both capital and liquidity, the relationship between the two also matters.Where a firm has little need of short-term funding to maintain its ongoing business, it isless susceptible to runs. Where, on the other hand, a firm is significantly dependent on suchfunding, it may need considerable common equity capital to convince market actors that it isindeed solvent. Similarly, the greater or lesser use of short-term funding helps define a firm’srelative contribution to the systemic risk latent in these markets.” - Remarks by Daniel K.Tarullo, Member of the Board of Governors of the Federal Reserve System, Peterson Institutefor International Economics, May 3, 2013.
1 Introduction
The main function of banks is to provide liquidity by offering funding (deposits) that ismore liquid than their asset holdings (Diamond and Dybvig (1983)). This liquidity mis-match, part of their business model, makes banks vulnerable to runs as creditors can de-mand immediate repayment when the bank faces asset shocks. The rationale for studyingthe solvency-liquidity nexus of banks is based on the literature explaining bank runs basedon the strength of the bank’s fundamentals. In Allen and Gale (1998), banking panics arerelated to the business cycle where creditors run if they anticipate that the bank’s asset val-ues will deteriorate. Similarly, Gorton (1988) shows that bank runs are systematic responsesto the perceived risk of banks.
Theoretical models on the two-way interaction between solvency and liquidity have beenmore recently developed. Diamond and Rajan (2005) show that bank runs, by making banksinsolvent, exacerbate aggregate liquidity shortages. In Rochet and Vives (2004), there is anintermediate range of the bank assets value for which the bank is still solvent but can fail iftoo many of its creditors withdraw, and the range of the interval decreases with the strengthof the bank’s fundamentals. Then, Morris and Shin (2008) explain that bank runs comefrom both the bank’s weak fundamentals and the “jitteriness” of its creditors. Therefore, thefailure region of the bank would be smaller if both the bank and its creditors held more cash.
An implication of this literature is that systemic risk is likely to play a key role in the
1
solvency-liquidity nexus through the liquidation costs caused by fire sales in a crisis. If thefirm fails in isolation, its illiquid assets can be liquidated for a price close to their value inbest use (Shleifer and Vishny (1992)).1 In a systemic crisis, however, potential buyers will beunable to find funding to buy the assets of the distressed firm. Creditors will consequentlyrun from banks that are vulnerable to an aggregate shock as they anticipate these banks willnot be able to repay them in a crisis.
While the solvency-liquidity nexus has been well studied theoretically in the economicliterature, the interaction between solvency and liquidity risks tends to be omitted in thenew capital and liquidity regulatory standards. The liquidity coverage ratio (LCR) of BaselIII imposes that financial firms hold a sufficient amount of high-quality liquid assets to covertheir liquidity needs over a month of stressed liquidity scenario.2 However, the liquidityneeds according to this standard are essentially a function of the funding mix of the bankand do not depend on other bank’s fundamentals, in particular, on its capital adequacy andasset risks. Similarly, the required capitalization of a bank in Basel III is not related to itsexposure to funding liquidity risk.3
The solvency-liquidity nexus of banks has also not been the center of empirical studiesinvestigating funding liquidity risk of the financial sector.4 In this paper, I fill this gap inthe literature and test whether the solvency-liquidity nexus of banks empirically holds byexamining the short-term balance sheet of 50 US bank holding companies over 2000-2013.
1Other fire-sale papers also relying on the Shleifer and Vishny (1992) insight include Allen and Gale(1998, 2000a,b, 2004); Acharya and Yorulmazer (2008); Acharya and Viswanathan (2011); Diamond andRajan (2005, 2011).
2Next to the LCR, Basel III also introduces a Net Stable Funding Ratio (NSFR). The NSFR is the ratioof available stable funding to required stable funding over a one year horizon. The required stable funding isdetermined based the institution’s assets and activities (Basel Committee on Banking Supervision (2011)).
3Funding liquidity risk is only likely to play a modest role via the interconnectedness measure usedto derive the additional capital requirement for globally systemically important financial institutions (G-SIFIs). The systemic importance measure is the equally-weighted average of the size, interconnectedness,lack of substitutes for the institution’s services, global activity and complexity (Basel Committee on BankingSupervision (2013b)). Interconnectedness is itself based on three indicator measures: intra-financial systemassets, intra-financial system liabilities, and securities outstanding.Alternatively, some supervisory stress test models explicitly feature funding liquidity feedbacks from the
deterioration of the banks’ fundamentals as in the risk assessment model for systemic institutions (RAMSI)of Aikman et al. (2009) used at the Bank of England.
4Related empirical studies include Das and Sy (2012) who document the trade-off between solvency andliquidity; banks with more stable funding and more liquid assets do not need as much capital to get thesame stock return. Gorton and Metrick (2012) find that increases in repo rates are correlated to higheraggregate counterparty risk, whereas increases in repo haircuts are correlated to higher uncertainty aboutcollateral values. Afonso et al. (2011) study the Fed funds market and find increased sensitivity to bank-specific counterparty risk during times of crisis (both in the amounts lent to borrowers and in the cost ofovernight funds).
2
Short-term debt mainly consists of Fed funds purchased and repurchase agreements (repos),uninsured deposits and other short-term borrowings. Short-term assets include cash, Fedfunds sold and reverse repos, and short-term debt securities.
The difference between short-term debt and short-term assets is used in this paper asa proxy for the exposure of a firm to funding liquidity risk. When liquidity conditions aretight, a financial firm faces the risk of not being able to roll-over its existing short-term debtand/or to raise new short-term debt. To survive a crisis, a firm needs a sufficient amountof liquid assets that can be converted into cash when the bank’s short-term funding startsdrying up. The gap between its short-term debt and short-term assets — the liquid assetshortfall — represents the amount of liquid assets that would be left if the bank lost itscomplete access to short-term funding (see Figure 1).
I test for the solvency-liquidity nexus using a fixed-effects panel vector autoregressive(VAR) model. In particular, I test for the interaction between solvency and liquidity usingseveral measures of solvency risk: regulatory capital ratios, market measures of risk (realizedvolatility, expected shortfall and market beta), and a measure of the expected capital short-fall (SRISK) of the bank under aggregate stress defined by Acharya et al. (2010, 2012);Brownlees and Engle (2011). According to SRISK, a firm is adequately capitalized to sur-vive a crisis if its ratio of market capitalization to total assets remains larger than 8% whenthe market index falls by 40% over the next six months. This measure is an alternative tothe capital shortfall estimates of stress tests that is purely based on publicly available marketdata (and therefore available at a higher frequency than stress tests outcomes).
I document four important results. First, I find that the bank’s capital shortfall understress (SRISK) determines how much short-term debt it can raise. This result supportsthe models of Allen and Gale (1998); Diamond and Rajan (2005), etc. explaining bank runsbased on the strength of the bank’s fundamentals. Conversely, the expected capital shortfallof a bank increases when the bank holds more short-term debt (has a larger exposure tofunding liquidity risk), in line with the introductory quote of D. Tarullo and some previousevidence that firms with more maturity mismatch have a larger contribution to systemic risk(Adrian and Brunnermeier (2010)). Figure 2 illustrates well the solvency-liquidity nexuswhere capital-constrained banks (i.e. banks with a positive SRISK) had a larger averageexposure to liquidity risk (measured by the difference between short-term debt and short-term assets) than adequately capitalized banks until 2011. The average liquidity shortfall ofcapital-constrained banks reached a maximum of $133 billion in the third quarter of 2007.This exposure made them particularly vulnerable to the sudden freeze of short-term funding
3
markets that followed.Second, I show that not all solvency risk measures predict the short-term debt level of
banks. The expected capital shortfall SRISK interacts well with the level of short-termfunding of the bank compared to other measures of solvency risk because (i) it is a measureof the bank’s exposure to aggregate risk, and (ii) it combines both book and market values.Relating to the model of liquidation costs of Shleifer and Vishny (1992), this result suggeststhat a bank with higher solvency risk in isolation does not necessarily get restricted accessto short-term funding. What matters most for the suppliers of short-term funding is thevulnerability of the bank to an aggregate crisis. When this crisis occurs, ’pure’ solvency risk(measured by the Tier 1 leverage ratio), amplified by market shocks, explains the bank’saccess to short-term funding.
Third, the stressed solvency risk measure interacts with the bank’s profitability (measuredby its net income divided by total assets) in determining its short-term balance sheet. Whilea more profitable bank has a larger access to short-term funding and does not hold as muchliquid assets, profitability does not have this beneficial effect on its short-term balance sheetwhen the bank is expected to be capital-constrained in a crisis. For example, the positivenet income of $2 billion of Citigroup in the third quarter of 2007 did not prevent the bankfrom losing 18% of its short-term funding (-$172 billion) the next quarter, as Citigroup wasalso highly undercapitalized according to SRISK ($51 billion expected capital shortfall in2007Q3).
Finally, out-of-sample forecasting results during the European sovereign debt crisis showthat the solvency-liquidity interaction helps improve the forecasts of the short-term balancesheet of banks. Omitting SRISK in the model increases the forecasting errors of the liquidasset shortfall considerably and particularly for capital-constrained banks.
Overall, the results of this paper suggest that the solvency-liquidity nexus should beaccounted for when designing liquidity and capital requirements, in contrast to Basel III reg-ulation where liquidity and solvency risks are treated separately. The paper gives empiricalsupport to the approach advanced by Tarullo (2013) to tie liquidity and capital requirementstogether by requiring banks with a large exposure to short-term funding to hold an addi-tional capital buffer. The liquid asset buffer of the LCR might be a sufficient requirementfrom a microprudential perspective. However, the sudden drop in short-term funding fora bank that has a perfectly maturity-matched securities book (including repos and reverserepos) may also result in fire sales and increases the risk of contagion by transferring fundingliquidity risk to the bank’s customers. The supplementary capital buffer is a preemptive
4
measure that would give the confidence to creditors to continue to provide funding to thebank in a period of aggregate stress.
The rest of the paper is structured as follows. In Section 2, I describe the short-termbalance sheet of banks and their solvency risk measures. I test the solvency-liquidity nexusin Section 3. I comment on the out-of-sample forecasting results in Section 4.
2 Short-term balance sheet and solvency risk measures
2.1 Long-term vs. short-term balance sheet and the liquid assetshortfall
The sample considered in this paper is a panel of 49 publicly traded US bank holdingcompanies (BHC) reporting their regulatory accounting data over 13 years from 2000Q1 until2013Q1 (i.e. 53 quarters). This sample of banks corresponds to the intersection betweenthe NYU Volatility Laboratory (V-Lab) sample for its global systemic risk analysis (thatwill be introduced in the next section) and the bank holding companies reporting under theFR Y-9C schedule (equivalent to the Call Reports of Condition and Income of commercialbanks). The names of the BHCs and their market capitalizations are reported in AppendixD.
I construct the short-term debt and short-term asset variables of these BHCs based onitems extracted from their FR Y-9C reports from the SNL Financial database. The short-term debt is constituted of uninsured time deposits of remaining maturity of less than a year,securities sold under agreements to repurchase (repos), Federal funds purchased, and otherborrowed money of remaining maturity of less than a year. The short-term assets includedebt securities of remaining maturity of less than a year, interest-bearing bank balances(cash), securities purchased under agreements to resell (reverse repos), and Federal fundssold. The components of short-term debt and short-term assets are described and illustratedin Appendix A.
As the panel data set is unbalanced, I will restrict the following analyses to a smallersample of 44 banks for which the time series dimension is larger than 30 observations.5 Itest the stationarity of the balance sheet quantities (in logarithms) in Appendix B usingthe panel unit root test robust to cross-sectional dependence of Pesaran (2007). This testindicates that the permanent impact of a shock on the size of a bank comes from shocks in
5This restriction excludes Goldman Sachs and Morgan Stanley from the sample as they obtained thestatus of bank holding company at the end of 2008.
5
the long-term balance sheet (where the unit root hypothesis is not rejected), whereas theshort-term balance sheet shocks revert to a trend level.6 This result is consistent with thelong-term balance sheet being the core business of the traditional bank that invests insureddeposits (part of the long-term debt) in loans (long-term assets).7
The evolution of the average balance sheet of banks is shown in Figure 3. The averagesize of the balance sheet (measured by total assets) triples (from $85 billion to $280 billion)over the sample period and follows an increasing trend in the long-term balance sheet. Overthis period, and particularly during the financial crisis, salient events include the acquisitionof out-of-sample banks by in-sample banks; Golden West Financial sold to Wachovia in May2006, Bear Stearns sold to J.P.Morgan in March 2008, Countrywide to Bank of Americain July 2008, Washington Mutual to J.P.Morgan and Merrill Lynch to Bank of America inSeptember 2008, and the acquisition of in-sample banks by other in-sample banks; NationalCity Corp. sold to PNC and Wachovia to Wells Fargo in the last quarter of 2008.
For the purpose of testing the solvency-liquidity nexus, this paper focuses on the short-term part of the balance sheet. The acquisition of two major investment banks (Bear Stearnsand Merrill Lynch) in 2008 brought a considerable amount of short-term debt and short-term assets in the banking sector (mostly in the form of repos and reverse repos). Theincrease in the average short-term balance sheet is considerable with the purchase of BearStearns (visible on J.P.Morgan’s balance sheet in 2008Q3). In comparison, the impact of theacquisition of Merrill Lynch (visible on Bank of America’s balance sheet in 2009Q1) on theaverage short-term balance sheet is attenuated as several large banks were losing a significantamount of short-term funding at that time.
In contrast to an overall increasing trend in short-term assets, the average short-termdebt slowed down in 2007Q3 with the first signs of a “run on repo” in August 2007 (Gortonand Metrick (2012)), visible on the short-term balance sheet of several large banks includingCitigroup that lost $172 billion (18%) of short-term debt from 2007Q3 to 2007Q4. Theaverage short-term debt reached a peak in the third quarter of 2008 (with the acquisition ofBear Stearns) and declined afterwards.
The gap between the short-term debt and short-term assets of a bank — its liquid assetshortfall — represents the amount of liquid assets that would be left if the bank lost itscomplete access to short-term funding.8 The average liquidity gap of the banking sector (also
6The trend stationarity of the short-term balance sheet allows estimating a dynamic panel data modeldirectly on the levels in Section 3 by applying standard estimation and inference techniques.
7The long-term debt (resp. assets) is the difference between total liabilities (resp. assets) and short-termdebt (resp. assets).
8Also note that short-term assets will serve in this paper as a proxy for liquid assets due to the lack of
6
shown in Figure 3) was the widest at the end of 2007 making banks particularly vulnerable tothe sudden freeze in short-term funding markets. The short-term funding freeze was furtheraccentuated with credit risk concerns at the end of 2008 with Lehman Brothers’ bankruptcyand the most negative average net income of banks over the sample period ($-850 million).
Since the financial crisis, the average liquid asset shortfall of banks has been decliningto become negative in 2011 (i.e., banks now hold more short-term assets than short-termdebt). Several circumstances explain the increase of banks’ stock of short-term assets. Afirst explanation is linked to the persistent effect of the financial crisis on the real economywhere the demand for loans has been slowly recovering and outpaced by deposit growth. Asa result, banks have been investing in securities and (profitable) treasury products.9 In orderto obtain secured short-term funding, banks also need to hold more short-term liquid assetsthan before due to stricter collateral requirements (higher haircuts). Then, higher liquidasset holdings by banks respond to precautionary concerns by banks (protecting againstanticipated interest rate increase) and the regulator. Banks are encouraged by regulationto hold more short-term liquid assets to comply with both liquidity requirements (BaselIII liquidity coverage ratio) and capital requirements (as holding short-term assets usuallyinvolves low regulatory capital requirements).
2.2 Solvency risk measures
2.2.1 Regulatory capital ratios
The regulator usually employs capital ratios to assess the solvency risk of a bank. Figure 4displays the average regulatory capital ratios: the Tier 1 common capital ratio (T1CR) andthe Tier 1 leverage ratio (T1LV GR). The Tier 1 common capital ratio is the ratio of Tier 1common equity to risk-weighted assets, whereas the Tier 1 leverage ratio is the ratio of Tier1 capital to total assets. The upward shift in regulatory capital ratios in the fourth quarterof 2008 indicates a healthier banking system and coincides with the launch on October14, 2008 of the Capital Purchase Program (CPP) and the Temporary Liquidity GuaranteeProgram (TLGP) under the Trouble Asset Relief Program (TARP). By purchasing assets
historical data for the assets included in the high-quality liquid assets (HQLA) definition of Basel III. Highquality liquid assets include cash, reserves at central banks, treasury bonds, and non-financial corporatebonds and covered bonds with the highest ratings. Additional assets like highly-rated RMBS, non-financialcorporate bonds and covered bonds with [A+, BBB-] rating, and common equity shares can be included inthe HQLA stock with the appropriate haircuts specified in the LCR revision of 2013 (Basel Committee onBanking Supervision (2013a)).
9“US banks brace for interest rate rises”, Financial Times, February 24, 2011. “Excess deposits demandnovel responses”, Financial Times, May 30, 2012.
7
and equity from troubled banks from October 2008 on, the TARP led to a significant increasein the average capital ratios. For example, Treasury bought $25 billion of preferred sharesof Citigroup in October 2008 and another $20 billion in November 2008 under the CPP.10
2.2.2 Expected capital shortfalls in a crisis
Acharya et al. (2012) define the systemic risk contribution of a firm i to the real economyat time t as “the real social costs of a crisis per dollar of capital shortage(t)× Probabilityof a crisis(t) × SRISKit”, where SRISKit represents the expected capital shortfall of thefirm in a crisis, i.e. when the market equity index drops by 40% over the next six months.In these market conditions, SRISK is based on the assumption that the book value of the(long-term) debt Dit of the bank will remain constant over the six-month horizon whileits market capitalization MVit will decrease by its six-month return in a crisis, called thelong-run marginal expected shortfall (LRMES). The expected capital shortfall in a crisisSRISK of bank i at time t is defined by
SRISKit = Et[k(Dit+h +MVit+h)−MVit+h|Rmt+h ≤ −40%] (1)
= kDit − (1− k) ∗MVit ∗ (1− LRMESit)
where Rmt+h is the return of the market index from period t to period t + h (h = 6months), k is the prudential capital ratio (8% for US financial firms), and LRMESit =
−Et(Rit+h|Rmt+h ≤ −40%). Compared to other market-based measures of systemic risk likethe CoVaR of Adrian and Brunnermeier (2010) or the Distress Insurance Premium (DIP)of Huang et al. (2012), an interesting feature of SRISK is that it is a function of size andleverage which are two characteristics that the regulator finds particularly relevant whenmeasuring solvency risk of banks. SRISK can be written as a function of size, leverage andrisk
SRISKit = MVit {k(Lvgit − 1)− (1− k)(1− LRMESit)} (2)
where Lvgit is the quasi-market leverage defined as the ratio of quasi-market assets to marketcapitalization (Lvgit = (MVit + Dit)/MVit). Therefore, the capital shortfall of a bank willbe large if the bank is large, highly leveraged and highly sensitive to an aggregate shock asmeasured by LRMESit.
These measures (SRISK and LRMES) are available from the V-Lab website developedat NYU Stern School of Business.11 In the global systemic risk analysis of V-Lab, LRMES
10See http://www.treasury.gov/initiatives/financial-stability/reports/Pages/TARP-Tracker.aspx.11See http://vlab.stern.nyu.edu/.
8
is extrapolated from its short-term counterpart MES representing the daily return of thebank conditional on a 2% decline in the daily return of a global market index. The MESis derived from a time-varying beta estimated with the Dynamic Conditional Beta model ofEngle (2012) that accounts for asynchronous trading around the world when measuring thecomovement of bank returns with a global market index.
By definition, SRISK can be negative when a bank is expected to hold a capital excess ina crisis. In Figure 4, we find two different regimes for the average SRISK of banks. SRISKwas indeed negative for most banks before 2007. The average SRISK was the lowest in thethird quarter of 2006 then started to increase in 2007. SRISK became positive in the fourthquarter of 2007 and reached a maximum average capital shortfall of $16 billion in the firstquarter of 2009. The average capital shortfall has remained positive since the financial crisisand bumped several times afterwards, in particular in the heat of the European sovereigndebt crisis in 2011.
3 Testing the solvency-liquidity nexus of banks
As liquidity risk concerns both sides of the balance sheet, I test the interaction betweensolvency risk and both the short-term debt and short-term assets. Panel unit root testsindicate that the variables yit = ln(STDebtit), zit = ln(STAssetsit), and the solvency riskmeasure SRISKit/TAit are trend stationary (see Appendix B). Therefore, the solvency-liquidity nexus is tested using a fixed-effects panel vector autoregressive (VAR) model forthe (K × 1) vector of endogenous variables wit
wit = αi + Φwit−1 + θt+ εit, t = 1, 2, ..., Ti, i = 1, 2, ..., N, (3)
where αi are bank dummies, θ is a trend parameter and Φ is a (K × K) matrix of VARparameters.12 Based on in-sample fit criteria, I augment the panel VAR process of eq. (3)to allow for heterogenous trend and heterogenous dynamic parameters
wit = αi + φi � wit−1 + θit+ δwit−1 + εit, (4)
where φi, θi are (K × 1) vectors of parameters specific to bank i, δ is a (K ×K) matrix ofparameters with zeros on the diagonal, and � is the Hadamard product.
12The parameters of eq. (3) are estimated by ordinary least squares. The bias of OLS parameter estimatesis likely to be small for the considered sample since the minimum size of the time series dimension for eachbank is 30 observations (i.e. Ti ≥ 30, ∀i).
9
The estimates of the interaction parameters (δ) of equation (4) are reported in Table1 where wit = (yit, zit, SRISKit/TAit)
′. This Table reveals the empirical solvency-liquiditynexus where banks with a larger expected capital shortfall find it more difficult to raiseshort-term funding; the estimates suggest that a positive unit shock on the ratio of SRISKto total assets produces a -1.102% shock on the short-term funding of the bank. On theother leg of the interaction, a 1% increase in the short-term debt of the bank increases itscapital shortfall ratio by 0.009. Therefore, the interaction between solvency and short-termdebt is asymmetric; higher solvency risk limits the access of the firm to short-term fundingbut a firm with more short-term debt has a higher risk of insolvency in a crisis.
This result supports the theoretical literature explaining bank runs based on the strengthof the banks fundamentals (Allen and Gale (1998); Diamond and Rajan (2005), etc.), anddescribing the interaction between liquidity and solvency problems of banks (Diamond andRajan (2005); Morris and Shin (2008); Rochet and Vives (2004)). The results also giveempirical support to the recent speeches by Carney (2013) and Tarullo (2013) explainingthat the repair of banks’ balance sheet (i.e. higher capital levels) gives the confidence toinvestors and creditors to continue to provide funding to banks.
From Table 1, we also note that short-term assets do not relate to the other variablesin the vector wit, suggesting that banks are not able to adjust their stock of short-termassets to solvency risk or short-term funding conditions in a timely fashion. It also reflectsa liquidity hoarding tendency of banks where banks prefer to sell long-term assets to repayshort-term creditors. Banks prefer to hold the short-term assets for precautionary reasonsor for investing in fire sale assets of other financial institutions that are expected to generatehigh future returns (Acharya et al. (2009)).
In the rest of this section, I test alternative solvency risk measures to predict the short-term balance sheet of banks in Section 3.1, for the interaction between profitability andsolvency risk in predicting the short-term balance sheet (Section 3.2), and for the robustnessof the solvency-liquidity nexus in Section 3.3.
3.1 Testing alternative solvency risk measures
I report the tests of alternative measures of solvency risk to predict the short-term balancesheet (yit and zit) in Table 2, controlling for the market-to-book ratio as the regressionincludes both accounting and market variables. The columns (1) to (6) show the individualimpact of each measure. From this Table, the regulatory capital ratios (T1CR and T1LV GR)do not appear to be related to either side of the short-term balance sheet. Market measures
10
of risk like the realized quarterly volatility is significant (at 5%) to predict short-term assetsbut this result does not hold in the regression including all solvency risk factors (column (7)).Then, the sensitivity of the bank’s return to market shocks measured by the the DynamicConditional Beta (DCB) of Engle (2012), and the contribution of the bank to systemic riskmeasured by the Delta CoVaR of Adrian and Brunnermeier (2010) are not significant driversof the short-term balance sheet either. When all solvency risk factors are included in theregression (column (7)), only SRISK per unit of asset and the market-to-book ratio aresignificant at the 1% level to predict the short-term debt level of banks.
The results of Table 2 suggest that not all solvency risk factors can predict the shocks inthe short-term balance sheet of banks. A bank with higher solvency risk in isolation doesnot necessarily get restricted access to short-term funding. However, banks lose short-termfunding when they are expected to be insolvent in a systemic crisis. An explanation for thisobservation is based on the liquidation costs of a firm’s illiquid assets in a crisis. Shleifer andVishny (1992) show that when a firm is individually in distress, its liquidation costs are notas high because the firm can find buyers in the same industry who value its illiquid assetsat a price close to their value in best use. In a crisis, however, the potential buyers in theindustry will likely also meet difficulties to find funding and will not be able to buy thoseassets. The firm will then have to sell its illiquid assets to less specialized buyers outside theindustry at a higher liquidation cost.
A bank that is expected to be insolvent in a crisis will be facing high liquidation costs andwill consequently not be able to raise cash. Creditors who anticipate this based on publiclyavailable data (as those used to derive SRISK) will run from the bank as they expect thebank will not be able to repay them. The liquidation costs during the 2008 financial crisiswere exacerbated by the huge gap between short-term assets and short-term debt observedin Section 2. As a result, banks had no choice but to sell illiquid assets to repay creditorswhen losing access to short-term funding.
In-sample fit criteria show the superiority of SRISK in Table 3 (first column) to predictshort-term funding; the adjusted R2 is 15.7% compared to an adjusted R2 around 11% forthe regressions with the alternative solvency risk measures of Table 2.13 In order to identifywhat works so well in SRISK for predicting the short-term funding of banks, Table 3 alsoreports the estimates of the different components of SRISK highlighted in eq. (2). TheTable shows that the improvement in in-sample fit rather comes from the ratio of market
13Note that all reported R2 are on the first differences (wit −wit−1). The R2 of levels (wit) are very high(around 90%) given the bank specific constant, trend and autoregressive parameters.
11
capitalization to total assets (MV/TA) than from the long-run marginal expected shortfall(LRMES) or the quasi-market leverage (Lvg). The main difference between Lvg and theMV/TA ratio is a different combination of book and market values; the ratioMV/TA is theproduct of the book leverage ratio (T1LV GRit) and the market-to-book ratio (MVit/BVit)
MVit
TAit
=BVit ∗
(MVit
BVit
)TAit
' T1LV GRit ∗(MVit
BVit
),
whereas Lvgit = 1 + Dit
MVitis not a function of the book leverage ratio. A potential argument
against using market values to test for the solvency-liquidity nexus is that market valuesalso incorporate information about liquidity risk. The results of Table 3 however suggestthat both the book solvency ratio — informing about pure solvency shocks — and themarket-to-book ratio — informing about how faster the market values fall compared to bookvalues — are important for predicting the short-term debt of banks. The ratio MV/TA ishighly correlated to the book leverage ratio (0.91) and less correlated to the market-to-bookratio (0.44); solvency risk amplified by market shocks explain banks’ access to short-termfunding and neither the market-to-book or the leverage ratio taken separately, nor theirlinear combination predict short-term funding.
The modest improvement in fit due to the downside risk of the bank LRMES (0.66%increase of adjusted R2 from column 4 to column 5, Table 3) is consistent with the sampleperiod that contains several episodes of market stress. In a crisis, all is already functionof the aggregate shock. However, measuring the downside risk is important preemptively;I find increasing out-of-sample forecasting errors when MV/TA is employed in the panelVAR instead of SRISK/TA for predicting the short-term balance sheet of banks during theEuropean sovereign debt crisis (especially with the dynamic forecasting exercise of Section4).
3.2 Interaction between solvency and profitability
In Perotti and Suarez (2011), both liquidity risk and profitability are increasing functionsof the short-term debt level of the bank. A bank will indeed demand more short-termfunding when it finds profitable investment opportunities. Its liquidity risk will also increaseas its short-term debt will be invested in long-term profitable assets. The impact of theprofitability of the bank measured by its net income divided by total assets is found to bepositive on short-term debt and negative on short-term assets in Table 4 (Panel A), butthese parameters are not significant at the 5% level.
12
The parameters of eq. (4) are however expected to vary with the state of the bank and/orthe aggregate liquidity conditions. In good times, short-term funding and short-term assetsare the result of management decisions and are driven by demand factors. As mentioned,banks with profitable opportunities will demand more short-term funding. In bad times,supply factors determine how much short-term debt a bank can raise and the short-termassets adjust accordingly. One way to disentangle supply and demand effects on the bankcharacteristics is to augment equation (4) with a state variable
wit = αi + φi � wit−1 + θit+ δwit−1 + γwit−1 ∗ st−1 + ωst−1 + εit (5)
where ω is a (K×1) vector of parameters, γ is a (K×K) matrix of parameters with zeros onthe diagonal, and the state variable st could be a bank characteristic or a common factor. Forexample, Cornett et al. (2011) use the TED spread (the difference between 3-month LIBORrate and T-bill rate) to reflect the change in the management of liquidity risk exposuresof banks during the financial crisis.14 In Table 4 (Panel B), I show that a good candidatefor the state variable is simply a dummy variable equal to one when SRISK is positive(sit = 1{SRISKit>0}), i.e. when the bank is expected to have a capital shortfall in a crisis.
This distinction between states where SRISK is positive or negative appears to beimportant when measuring the effect of the profitability of the bank on its short-term balancesheet. Indeed, a bank with a higher net income has a larger access to short-term fundingwhile it does not hold as much liquid assets. In Table 4, this beneficial effect of the bank’sprofitability on its short-term balance sheet appears to be true only when the bank’ SRISKis negative, i.e. when the bank is adequately capitalized to survive a crisis (sit = 0). Whenthe bank is expected to be capital-constrained in a crisis (sit = 1), the effect of profitabilityon its balance sheet disappears (δ + γ ' 0), and only supply factors predict the short-termdebt of the bank.
An interesting observation is that the contrasting impact of profitability on the short-termbalance sheet can be reproduced when I allow for parameter breaks over time
wit = αi + φi � wit−1 + θit+ δwit−1 + δcwit−1 ∗ ct + δpcwit−1 ∗ pct + ωcct + ωpcpct + εit (6)
where ct and pct are dummy variables indicating whether the quarter t belongs to the financialcrisis period (2007Q1-2009Q4) or the post crisis period (2010Q1-2013Q1) respectively. InTable 11 (Appendix C), we observe that the impact of the net income on the short-term
14The TED spread is however not significant to predict the short-term balance sheet for the sampleconsidered in this paper (cf.. Section 3.3.2).
13
balance sheet also disappeared during the financial crisis (δ + δc ' 0). This Table furthershows that SRISK was significant only during the financial crisis (when there was actualliquidity stress). The results tend to confirm the interpretation of SRISK as a supply factorfor short-term funding, and of the net income as a demand factor when the firm is adequatelycapitalized.
Based on the estimates of the panel VAR, I can derive the impulse response functions(IRF) of wit where wit = (yit, zit, NetIncomeit/TAit, SRISKit/TAit)
′. I collect the orthogo-nalized shocks of the VAR from a Cholesky decomposition of the covariance matrix of εit ofeq. (4), with the ordering given by [NetIncomeit/TAit → SRISKit/TAit → yit → zit]. Thisordering is motivated by the observation that exogenous shocks first impact a firm via itsactivities and is translated into the net income, markets react by adjusting their assessmentof the capital shortfall SRISK, this in turn affects how much funding the bank can accessand the short-term assets adjust in consequence.
Due to the heterogenous autoregressive parameters of eq. (4), firms will have heteroge-neous responses to sigma shocks. In Figure 5, I show the median impulse response functionto orthogonalized sigma shocks between the 25% and the 75% IRF quantiles to assess theheterogeneity in impulse responses across firms. In general, the IRF that concerns short-termfunding are more heterogenous (impact of other variables shocks on short-term debt and im-pact of short-term debt shocks on other variables). The range between impulse responsesquantiles is also wider for the interaction between SRISK and the short-term balance sheet.For some banks, it takes three years for the impact of SRISK shocks on short-term fundingto vanish whereas the solvency shocks of other banks have a more definitive impact on theirshort-term funding. Then, the impulse response functions well illustrate previous findingson the asymmetric impact of shocks between SRISK and the short-term balance sheet, andbetween the net income and the short-term debt.
The impact of orthogonalized sigma shocks will be different when I differentiate betweencapital-constrained vs. adequately capitalized banks based on the parameters of eq. (5).Figure 6 shows the gap in median impulse response functions between adequately capitalizedversus capital-constrained firms. The solvency-liquidity nexus appears to be exacerbatedfor capital-constrained banks; the impact of solvency shocks on short-term funding doublescompared to adequately capitalized banks, while the response of short-term debt to shocks inother bank characteristics (net income and short-term assets) is less important and vanishesmore rapidly. For capital-constrained banks, only the solvency-liquidity nexus appears toexplain the short-term balance sheet.
14
3.3 Robustness of the solvency-liquidity nexus
3.3.1 Robustness to TARP
On October 14, 2008, the U.S. government announced a series of measures — the TroubledAsset Relief Program (TARP) — to restore financial stability. Under the TARP, the Trea-sury Department launched the Capital Purchase Program (CPP) and the Federal DepositInsurance Corporation launched the Temporary Liquidity Guarantee Program (TLGP). Un-der the CPP, Treasury injected $205 billion capital into banks by buying warrants, commonshares, and preferred shares.15 Under the TLGP, the FDIC allowed financial institutions toretain and raise funding by giving a guarantee on existing noninterest-bearing transactionaccounts and certain newly issued senior unsecured debt. Data on the amount and maturityof total unsecured debt issued by banks and guaranteed by the FDIC are publicly available.16
It is possible to derive the amount of short-term debt a bank would have had if it did nothave access to TLGP funding. The solvency-liquidity nexus estimates hardly change whenTLGP funding is not taken into account. It is however impossible to project this scenarioon the other variables of the panel VAR as it requires to know where TLGP funding wasinvested and how markets would have reacted in this scenario.
If we assume that banks received help from the TARP when they actually needed it,the amount of CCP capital injected and the amount of TLGP funding received are realizedmeasures of the bank’s capital shortfall and liquidity shortfall, respectively. The largestbanks received the largest injections of capital and liquidity. Looking at data from the secondquarter of 2008, I test different bank characteristics and risk measures to explain their capitaland liquidity shortfalls divided by their total assets. In Table 5, I report the estimates ofcross-sectional regressions for a sample of 17 banks that received both capital and liquidityinterventions. It appears that the regulatory capital ratios are the most important factorsexplaining government interventions. After controlling for the size, we observe that banksthat received help from the CPP and banks that received help from the TLGP have differentprofiles. Banks that received government secured debt had low Tier 1 leverage ratios (ratioof Tier 1 capital to total assets), whereas banks that received capital injections had low Tier1 Common capital ratios (ratio of Tier 1 Common capital to risk-weighted assets). Bankswith high liquidity shortfalls in 2008Q4 had a large short-term balance sheet, high cost offunding, and high market beta (or LRMES) in 2008Q2. Conversely, banks with high capital
15http://www.treasury.gov/initiatives/financial-stability/TARP-Programs/bank-investment-programs/cap/Pages/overview.aspx
16See http://www.fdic.gov/regulations/resources/TLGP/index.html
15
shortfalls were more traditional banks with a large long-term balance sheet (deposits andloans), high yields on earning assets, and low market beta (among the banks that receivedgovernment interventions).
3.3.2 Common factors
The short-term balance sheets of firms are expected to co-move according to the aggregateliquidity conditions. To capture these common effects, I consider the macroeconomic andfinancial factors that are used in Fontaine and Garcia (2012) to relate to their factor mea-suring the value of funding liquidity, and will test the robustness of the solvency-liquiditynexus to these factors in the next section. The sensitivity of the short-term balance sheet(and its covariates) to the common factors is tested in
wit = αi + φi � wit−1 + θit+ β′ft−1 + εit (7)
where wit = (yit, zit, NetIncomeit/TAit, SRISKit/TAit)′, and ft is a vector of common fac-
tors. Note that common factors do not necessarily need to be lagged but this allows for thederivation of one-step ahead forecasts for wit without specifying a model for the commonfactors. The estimated parameters of common factors of eq. (7) are reported in Table 6.17
Interest rates are expected to play an important role on the short-term balance sheet.Three factors related to interest rates are considered; the level of interest rates is capturedby the Fed funds rate, the difference between long-term and short-term rates is measuredby the slope factor of the Treasury yield curve, and the TED spread reflects the perceivedcounterparty risk of interbank loans compared to Treasury loans. The TED spread is usuallyreferred as an aggregate funding liquidity risk factor (Cornett et al. (2011); Fontaine andGarcia (2012)). In the sample considered, the TED spread is not significant to explain theshort-term balance sheet directly but has a negative impact on the profitability of banks anda positive impact on their solvency risk (measured by SRISK).
The Treasury slope factor measures the difference between long-term and short-terminterest rates. A steeper term structure indicates higher profitability of investing short-termfunding in long-term assets (Fontaine and Garcia (2012)). This factor also reflects businesscycles and could be interpreted as a demand factor for liquidity. It is therefore not surprisingthat short-term debt increases with a steeper slope of the Treasury yield curve.
17Data sources of common factors: Federal Reserve Board Selected Interest Rates - H.15 (Fed fund rate);FRB Money Stock Measures - H.6 (M2 money supply growth); FRB Financial Accounts of the United States- Z.1 (MMMF flows, mortgage growth); Department of the Treasury (Treasury yield curves); Bloomberg(VIX).
16
The positive and significant coefficient of the Fed funds rate on short-term debt is moresurprising and possibly reflects an endogenous response of the Federal Reserve to fundingconditions during the financial crisis. The impact of interest rates on bank’s funding mayhowever be subtler. Diamond and Rajan (2005) explain that higher interest rates do notalways lead to lower excess demand for liquidity because of the effect of bank failures. Higherinterest rates cause more banks to become insolvent and run (because of decreasing assetsvalue). The excess demand will increase with interest rates if by failing, banks absorbmore liquidity than when solvent. Through these two channels (Fed interventions and firmsfailures), solvency risk has an aggregate endogenous feedback on the level of interest rates.
Mortgage growth (MTG) increases the demand for short-term debt. MTG is referredin Fontaine and Garcia (2012) as a factor exclusively affecting the demand for liquidityby increasing the pool of illiquid assets in the economy. Other considered factors includeflight-to-quality variables related to Money Market Mutual Funds (MMMF). The growthin MMMF assets (MMG) increases the supply of funding to banks via the shadow bankingsector (Adrian and Shin (2009); Fontaine and Garcia (2012)), but short-term funding supplydecreases when MMMF assets are allocated to safer assets like time deposits (MMA1) orgovernment-sponsored securities (MMA2).
The coefficient associated with MMA1 is negative and significant at the 1%. This resultcould, however, simply reflect the increase of the deposit insurance limit of the FederalDeposit Insurance Corporation (FDIC) in 2008Q4. Acharya and Mora (2013) document theshift from time deposits and debt issued by banks (and MMA1) to government-sponsoredsecurities (and MMA2), and the “liquidity reversal” in 2008Q4 where MMA1 started toincrease again. When the FDIC deposit insurance limit increased from 100K to 250K in thefourth quarter of 2008, uninsured deposits included in short-term debt shifted to the long-term part of the balance sheet. Therefore, the negative impact of MMA1 on banks’ short-term debt partly corresponds to the reallocation in 2008Q4 of some previously uninsuredtime deposits to the long-term debt within banks’ balance sheets.
We also note the positive coefficient of the VIX as banks’ exposure to short-term debt wasthe highest when the VIX peaked during the financial crisis. Finally, short-term assets arenot sensitive to any of the considered factors. While the level of short-term assets adjuststo shocks in other parts of the balance sheet, it is not directly affected by financial andmacroeconomic conditions.
17
3.3.3 Robustness of the solvency-liquidity nexus to common factors
I test the robustness of the solvency-liquidity nexus to the presence of common factors in
wit = αi + φi � wit−1 + θit+ λ′git−1 + β′ft−1 + εit (8)
where wit = (yit, zit, NetIncomeit/TAit, SRISKit/TAit)′, git is a ((2 ∗ K + 1) × 1) vector
stacking wit, wit ∗ sit and sit in a single column, λ is a ((2 ∗K + 1)× 1) vector containing theδ, γ and ω parameters, and ft is the vector of macroeconomic and financial factors identifiedin Section 3.3.2.
Chudik and Pesaran (2013) propose an alternative modeling strategy based on the Com-mon Correlated Effects (CCE) of Pesaran (2006) where the unobserved common factors areproxied by the cross-sectional averages of the dependent variable and the regressors
wit = αi + φi � wit−1 + θit+ λ′git−1 +1∑
l=0
ϕ′lwt−l + κ′gt−1 + εit (9)
where wt−l = N−1∑N
i=1wit−l and gt = N−1∑N
i=1 git.The estimation results of eq. (8) and eq. (9) are reported in Table 7. The fit improves
considerably when common factors are included. The best in-sample performance is foundwith the CCE model for all elements of wit. However, the CCE model counts a contempora-neous factor (average of the dependent variable) while the model with macro and financialfactors only includes lagged factors. The macro-financial model is therefore more convenientfor forecasting and the loss of in-sample fit is relatively small compared to the CCE model.
The solvency-liquidity nexus holds when I control for cross-sectional dependence. Theinteraction term between the profitability and SRISK is however not as important (notsignificant at the 5%). The impulse response functions do not qualitatively change eitherwhen macroeconomic and financial factors are considered.
3.3.4 Short-term debt components and long-term leverage
The different components of short-term debt (repos, uninsured deposits, commercial papers,etc.) have very different characteristics and may not react to solvency risk with the samemagnitude. Table 12 (Appendix C) reports the parameter estimates of eq. (4) where thedependent variable in each column is a different component (in logarithm) of the short-termdebt available from FR 9-YC reports. SRISK predicts most of the components of theshort-term debt; it is significant at the 1% level for wholesale funding (Fed funds, repos,
18
and commercial papers), and at the 5% level for retail funding (uninsured time deposits andforeign office deposits).
The expected capital shortfall SRISK is only related to the short-term part of thebalance sheet and does not predict long-term leverage. Table 8 shows that the long-termbalance sheet is not related to SRISK. The long-term debt only reacts to short-term assetsand the change in long-term assets.
Other robustness checks (not reported in this paper) show that the interaction betweensolvency and liquidity remains with homogenous dynamic parameters (φi = φ, ∀i), homoge-nous trend parameters (θi = θ, ∀i), without trend (θi = 0, ∀i), and when a break in 2008Q4is included in the trend. These results tend to confirm the robustness of the solvency-liquidity nexus. In the next section, I test for the out-of-sample forecasting performance ofthe solvency-liquidity nexus in predicting the short-term balance sheet of banks.
4 Forecasting the short-term balance sheet
To test for the out-of-sample predictive performance of the solvency-liquidity nexus, I con-duct two forecasting exercises. Both exercises are based on a fixed estimation period from2000Q1 to 2010Q4 to forecast the balance sheet of banks over the four quarters of 2011. Theinformation is updated each quarter in the one-step ahead forecasts (wit+1|t), while there isno information update in the dynamic forecasts (wit+h|t). The out-of-sample period corre-sponds to the European sovereign debt crisis, funding conditions were not as tight as duringthe financial crisis in the US but there is a total decline of $161 billion in short-term fundingof US banks during this period.
The root-mean square forecasting error (RMSFE) of the one-step ahead forecasting exer-cise are reported in Table 9. In this Table, I report the RMSFE of the short-term debt andshort-term assets individually (Panel A and B), as well as the RMSFE of their difference(Panel C). As already mentioned, the liquid asset shortfall is a measure of the exposure ofbanks to funding liquidity risk; the wider the gap in the short-term balance sheet, the morevulnerable the bank to runs. As this paper studies the liquidity-solvency nexus of banks,I also report the RMSFE of this liquid asset shortfall for capital-constrained (Panel D) vs.adequately capitalized banks (Panel E).
Four models are considered: a univariate autoregressive model (AR), the panel VARmodel of eq. (4) (VAR), the panel VAR model of eq. (5) that allows for the interaction ofbank characteristics with the state variable sit = 1{SRISKit>0} (INT), and the model including
19
all these features together with the macroeconomic and financial factors (eq. (8)) (CF).The assumption on the trend appears to be the most important model characteristic to
impact forecasting errors. To check for the robustness of the forecasting results, I report theRMSFE of these models for different trend assumptions (heterogeneous trends, homogenoustrend, no trend, and a break in the homogenous trend in 2008Q4).
For the one-step ahead forecasts, the best model is the panel VAR model that accountsfor the interaction of bank characteristics with SRISK (INT), and that assumes a breakin the trend in the fourth quarter of 2008. When the trend parameters are constant overtime, the model with common factors (CF) performs the best for the liquid asset shortfall ascommon factors reflect the changing aggregate funding conditions after the financial crisis.In the last three columns of Table 7, I report the increase in RMSFE when a particular bankvariable is not included in the panel VAR model.This Table shows that omitting SRISKincreases the forecasting errors of the liquid asset shortfall considerably, and particularly forcapital constrained banks during 2011. However, the panel VAR model or the interactionwith solvency risk (INT) does not improve the forecasts of adequately capitalized banks.
I obtain very similar results for the dynamic forecasts and therefore do not report theirRMSFE. Note that the RMSFE of dynamic forecasts are larger compared to the errors ofone-step ahead forecasts due to the absence of information updates over the forecastinghorizon. The model with interaction with SRISK (INT) and a break in the trend afterthe financial crisis is also the preferred model according the RMSFE of dynamic forecasts.The cross-sectional average dynamic forecasts obtained with this model for the short-termbalance sheet levels and flows over 2011Q1-2013Q1 are illustrated in Figure 7. It turns outthat the model is outstanding at forecasting short-term financing flows but does a less goodjob at forecasting short-term asset flows, which are not sensitive to the factors considered inthe model.
In Figure 8, I show the average dynamic forecasts of the liquid asset shortfall across allbanks, as well as for the subsamples of capital-constrained vs. adequately capitalized banks.As mentioned in the introduction, the liquid asset shortfall of capital-constrained banksspiked in the first quarters of 2007 and suddenly dropped afterwards due to the suddenfreeze of short-term funding markets. In the first quarter of 2011, the average liquid assetshortfall of capital-constrained banks became negative; capital-constrained firms had lessexposure to funding liquidity risk than adequately capitalized banks for the first time overthe sample period. The model predicts this reversal in the solvency-liquidity nexus andpredicts well the average excess of liquidity of capital-constrained banks during this period.
20
5 Conclusion
This paper reveals the empirical solvency-liquidity nexus of banks. While the interactionbetween solvency and liquidity has been well studied in the theoretical economic literature,this relationship tends to be omitted in the new capital and liquidity regulatory standardsintroduced under Basel III. In this paper, I test the solvency-liquidity nexus by examiningthe short-term balance sheet and the solvency risk measures of a sample of US bank holdingcompanies over 2000-2013.
I find that the expected capital shortfall of a bank in a crisis (SRISK) predicts how muchshort-term funding the bank has access to. Conversely, when the bank holds more short-termdebt, its risk of insolvency in a crisis increases. This result appears to be strong under manyrobustness checks and supports the theoretical models of the interaction between solvencyand liquidity risks and its amplification (aggregate) effects leading to systemic risk.
Importantly, not all solvency risk measures predict the bank’s access to short-term debt.The expected capital shortfall SRISK interacts well with the level of short-term funding ofthe bank compared to other solvency risk measures because (i) it is a measure of the bank’sexposure to aggregate risk, and (ii) it combines both book and market values. Suppliers ofliquidity are mostly concerned with the vulnerability of the bank to an aggregate crisis dueto the high liquidation costs the distressed bank will face in the presence of fire sales. Whenthe crisis happens, ’pure’ solvency risk (measured by the Tier 1 leverage ratio) amplified bymarket shocks explains the bank access to short-term funding.
The expected capital shortfall of the bank under stress also interacts with its profitabilityin determining its short-term balance sheet. While a profitable bank gets a larger accessto short-term funding and does not hold as much liquid assets, the impact of the bank’sprofitability on its liquidity profile tends to disappear when the bank is expected to becapital-constrained in a crisis.
The solvency-liquidity nexus provides useful information for forecasting the short-termfinancing flows during 2011 (European sovereign debt crisis). I show that the forecastingerrors of the liquid asset shortfall of banks increase considerably when the stressed solvencyrisk measure is not included in the regression.
Overall, the results of this paper suggest that the solvency-liquidity nexus should beaccounted for when designing liquidity monitoring tools and prudential requirements. Thisfinding contrasts with Basel regulation where solvency and liquidity risks are treated sepa-rately and gives empirical support for an additional capital requirement for banks with largeexposure to short-term funding.
21
References
Acharya, V., C. Brownlees, R. Engle, F. Farazmand, and M. Richardson (2010). Measuringsystemic risk. In V. Acharya, T. Cooley, M. Richardson, and I. Walter (Eds.), RegulatingWall Street: The Dodd-Frank Act and the New Architecture of Global Finance, Chapter 4.John Wiley & Sons, Ch. 4.
Acharya, V., R. Engle, and M. Richardson (2012). Capital shortfall: a new approach torankings and regulating systemic risks. American Economic Review Papers and Proceed-ings 102:3, 59–64.
Acharya, V. and N. Mora (2013). A crisis of banks as liquidity providers. Journal ofFinance forthcoming.
Acharya, V., H. Shin, and T. Yorulmazer (2009). Endogenous choice of bank liquidity: therole of fire sales. Bank of England Working Paper no. 376.
Acharya, V. and S. Viswanathan (2011). Leverage, moral hazard, and liquidity. Journal ofFinance 66, 99–138.
Acharya, V. and T. Yorulmazer (2008). Cash-in-the-market pricing and optimal resolutionof bank failures. Review of Financial Studies 21, 2705–2742.
Adrian, T. and M. Brunnermeier (2010). CoVaR. Federal Reserve Bank of New York StaffReports No. 348.
Adrian, T. and H. Shin (2009). Money, liquidity, and monetary policy. NY Fed Staff Reportno. 360.
Afonso, G., A. Kovner, and A. Schoar (2011). Stressed, not frozen: The federal funds marketin the financial crisis. Federal Reserve Bank of New York Staff Reports, no. 437.
Aikman, D., P. Alessandri, B. Eklund, P. Gai, S. Kapadia, E. Martin, N. Mora, G. Sterne,and M. Willison (2009). Funding liquidity risk in a quantitative model of systemic stability.Bank of England Working Paper No. 372.
Allen, F. and D. Gale (1998). Optimal financial crises. Journal of Finance 53:4, 1246–1284.
Allen, F. and D. Gale (2000a). Financial contagion. Journal of Political Economy. 108, 1–33.
22
Allen, F. and D. Gale (2000b). Optimal currency crises. Carnegie-Rochester ConferenceSeries on Public Policy 53, 177–230.
Allen, F. and D. Gale (2004). Financial intermediaries and markets. Econometrica 72:4,1023–1061.
Basel Committee on Banking Supervision (2011, June). Basel III: A global regulatory frame-work for more resilient banks and banking systems. Bank for International Settlements.
Basel Committee on Banking Supervision (2013a, January). Basel III: The Liquidity Cov-erage Ratio and Liquidity Risk Monitoring Tools. Bank for International Settlements.
Basel Committee on Banking Supervision (2013b, July). Global systemically importantbanks: updated assessment methodology and the higher loss absorbency requirement.Bank for International Settlements.
Brownlees, C. and R. Engle (2011). Volatility, correlation and tails for systemic risk mea-surement. NYU Working Paper.
Carney (2013). Crossing the threshold to recovery. Speech August 28, 2013.
Chudik, A. and H. Pesaran (2013). Common correlated effects estimation of heterogeneousdynamic panel data models with weakly exogenous regressors. Cambridge Working Papersin Economics no 1317.
Cornett, M., J. McNutt, P. Strahan, and H. Tehranian (2011). Liquidity risk managementand credit supply in the financial crisis. Journal of Financial Economics 101:2, 297–312.
Das, S. and A. Sy (2012). How risky are banks’ risk-weighted assets? Evidence from thefinancial crisis. IMF Working Paper WP/12/36.
Diamond, D. and P. Dybvig (1983). Bank runs, deposit insurance, and liquidity. Journal ofPolitical Economy 91:3, 401–419.
Diamond, D. and R. Rajan (2005). Liquidity shortages and banking crises. Journal ofFinance 60:2, 615–647.
Diamond, D. and R. Rajan (2011). Fear of fire sales, illiquidity seeking, and credit freezes.Quaterly Journal of Economics 126:2, 557–591.
Engle, R. (2012). Dynamic conditional beta. NYU Working Paper.
23
Fontaine, J.-S. and R. Garcia (2012). Bond liquidity premia. The Review of FinancialStudies 25:4, 1207–1254.
Gorton, G. (1988). Banking panics and business cycles. Oxford Economic Papers 40:4,751–781.
Gorton, G. and A. Metrick (2012). Securitized banking and the run on repo. Journal ofFinancial Economics 104:3, 425–451.
Huang, X., H. Zhou, and Z. Haibin (2012). Systemic risk contributions. Journal of FinancialServices Research 42, 55–83.
Morris, S. and H. Shin (2008). Financial regulation in a system context. Brookings Paperson Economic Activity 2008, 229–261.
Perotti, E. and J. Suarez (2011). A Pigovian approach to liquidity regulation. Journal ofInternational Central Banking 7:4, 3–39.
Pesaran, H. (2006). Estimation and inference in large heterogeneous panels with a multifactorerror structure. Econometrica 74:4, 967–1012.
Pesaran, H. (2007). A simple panel unit root test in the presence of cross-section dependence.Journal of Applied Econometrics 22, 265–312.
Rochet, J.-C. and X. Vives (2004). Coordination failures and the lender of last resort: wasBagehot right after all? Journal of the European Economic Association 2:6, 1116–1147.
Shleifer, A. and R. Vishny (1992). Liquidation values and debt capacity: A market equilib-rium approach. Journal of Finance 47:4, 1343–1366.
Tarullo, D. (2013). Evaluating progress in regulatory reforms to promote financial stability.Speech May 3, 2013.
24
Dep. variable: yit zit (SRISK/TA)it(SRISK/TA)it−1 -1.120** 0.074
(0.244) (0.114)
zit−1 -0.040 -0.001(0.023) (0.002)
yit−1 -0.003 0.009**(0.022) (0.002)
R2 (%) 20.811 22.157 15.151Adj. R2 (%) 15.430 16.868 9.429
Table 1: Testing the solvency-liquidity nexus. Estimates from pooled OLS regressionwith bank dummies, time trends, and heterogeneous AR parameters. Dependent variables:yit = ln(STDebtit), zit = ln(STAssetsit), (SRISK/TA)it = SRISKit/TotalAssetsit. Ro-bust standard errors in parentheses. * significant parameter at 5%; ** at 1%. Sample: 2107panel obs. over 2000Q1-2013Q1 (unbalanced), 44 banks. SRISK is the expected capitalshortfall of the bank in a crisis.
25
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Dep.variab
le:
y it
z it
y it
z it
y it
z it
y it
z it
y it
z it
y it
z it
y it
z it
T1C
Rit−
10.301
-0.070
0.041
-0.089
(0.337)
(0.134)
(0.552)
(0.360)
T1L
VGR
it−
10.508
-0.074
-1.148
-0.142
(0.491)
(0.305)
(0.875)
(0.832)
RealV
olit−
1-0.438
0.919*
2.286*
0.151
(0.412)
(0.450)
(1.119)
(1.491)
ES i
t−1
-0.250
0.412
-0.697*
0.364
(0.159)
(0.203)
(0.397)
(0.644)
DCB
it−
1-0.046
0.030
-0.021
0.002
(0.027)
(0.034)
(0.028)
(0.040)
4CoV
aRit−
1-0.402
0.188
0.094
0.102
(0.761)
(0.784)
(0.750)
(0.801)
(SRISK/T
A) i
t−1
-1.579**
-0.078
(0.072)
(0.141)
MB
it−
10.042
-0.014
0.042
-0.014
0.035
-0.001
0.033
0.000
0.036
-0.011
0.041
-0.015
-0.053**
-0.005
(0.026)
(0.017)
(0.025)
(0.017)
(0.027)
(0.017)
(0.026)
(0.017)
(0.025)
(0.019)
(0.024)
(0.017)
(0.017)
(0.022)
R2(%
)16.621
22.196
16.604
22.192
16.581
22.393
16.623
22.424
16.645
22.235
16.542
22.194
21.461
22.428
Adj.
R2(%
)10.955
16.909
10.937
16.905
10.913
17.119
10.957
17.153
10.980
16.951
10.871
16.907
15.868
16.904
Table2:
Testingalternativesolven
cyrisk
measures.
Estim
ates
from
pooled
OLS
regression
withba
nkdu
mmies,
timetrends
andheterogeneou
sAR
parameters.
Dep
endent
variab
les:y i
t=
ln(STDebt i
t),z i
t=
ln(STAssets i
t).T1C
R:
Tier1Com
mon
Cap
ital
Ratio,T
1LVGR:T
ier1Le
verage
Ratio,R
ealVol:Realized
volatility,
DCB:D
ynam
icCon
dition
alBeta(E
ngle
(201
2)),SRISK/TA
=SR
ISK/T
otal
Assets,
MB:m
arketto
book
equity
ratio.
Rob
uststan
dard
errors
inpa
rentheses.
*sign
ificant
parameter
at5%
;**
at1%
.Sa
mple:
2107
panelob
s.over
2000
Q1-20
13Q1(unb
alan
ced),44
bank
s.
26
(1)
(2)
(3)
(4)
(5)
(6)
Dep.variab
le:
y it
z it
y it
z it
y it
z it
y it
z it
y it
z it
y it
z it
(SRISK/T
A) i
t−1
-1.439**
0.010
(0.105)
(0.110)
LRMES i
t−1
-0.162
0.205
-0.080
0.195
(0.096)
(0.111)
(0.110)
(0.117)
Lvg i
t−1
-0.002
0.001
-0.002
0.000
(0.001)
(0.001)
(0.001)
(0.001)
(MV/T
A) i
t−1
0.930**
-0.002
0.925**
0.002
(0.051)
(0.049)
(0.052)
(0.046)
(SMV/T
A) i
t−1
1.369**
-0.021
(0.080)
(0.116)
MB
it−
1-0.048**
-0.014
0.032
-0.002
0.032
-0.007
-0.050**
-0.014
-0.051**
-0.013
-0.060
-0.001
(0.016)
(0.022)
(0.025)
(0.019)
(0.027)
(0.019)
(0.019)
(0.021)
(0.016)
(0.022)
(0.020)
(0.024)
R2(%
)21.110
22.191
16.714
22.443
16.701
22.254
20.725
22.191
21.338
22.192
20.931
22.446
Adj.
R2(%
)15.749
16.904
11.055
17.173
11.041
16.971
15.338
16.903
15.993
16.905
15.473
17.092
Table3:
TestingSRISK
compon
ents.Estim
ates
from
pooled
OLS
regression
withba
nkdu
mmies,
timetrends
and
heterogeno
usAR
parameters.
Dep
endent
variab
les:y i
t=
ln(STDebt i
t),z
it=
ln(STAssets i
t).MV:m
arketcapitalization,
TA:total
assets,R
ealVol:Realized
volatility,SRISK/TA
=SR
ISK/T
otal
Assets,LR
MES:
Long
-Run
Margina
lExp
ected
Shortfall,Lv
g:qu
asi-m
arketleverage,M
B:m
arketto
book
equity
ratio,
SMV/T
A=
MV*(1-LR
MES)/T
A.R
obuststan
-da
rderrors
inpa
rentheses.
*sign
ificant
parameter
at5%
;**
at1%
.Sa
mple:
2107
panelob
s.over
2000
Q1-20
13Q1
(unb
alan
ced),4
4ba
nks.
27
Pan
elA
:N
oin
tera
ctio
nw
iths i
t=
1 {S
RIS
Kit
>0}
Pan
elB
:In
tera
ctio
nw
iths i
t=
1 {S
RIS
Kit
>0}
Dep
.va
riab
le:
y it
z it
(NI/
TA
) it
(SR
ISK
/TA
) it
y it
z it
(NI/
TA
) it
(SR
ISK
/TA
) it
(SR
ISK
/TA
) it−
1-1
.063
**-0
.028
-2.2
25**
-0.9
35**
-0.1
20-1
.681
**(0
.245
)(0
.118
)(0
.301
)(0
.261
)(0
.101
)(0
.080
)(S
RIS
K/T
A) i
t−1∗
s it−
1-0
.408
1.75
7*-5
.871
**(0
.751
)(0
.767
)(1
.767
)
(NI/
TA
) it−
12.
354
-4.2
28-1
.217
**9.
704*
*-7
.944
*-2
.809
**(2
.278
)(2
.331
)(0
.389
)(3
.290
)(3
.716
)(0
.935
)(N
I/TA
) it−
1∗
s it−
1-9
.902
*6.
315
2.13
4*(4
.396
)(5
.183
)(0
.937
)
zit−
1-0
.038
-0.0
15-0
.002
-0.0
33-0
.035
-0.0
01(0
.023
)(0
.020
)(0
.002
)(0
.022
)(0
.022
)(0
.003
)z
it−
1∗
s it−
1-0
.021
*0.
078*
-0.0
008
(0.0
08)
(0.0
32)
(0.0
02)
yit−
1-0
.004
-0.0
67**
0.00
8**
-0.0
02-0
.031
0.00
8**
(0.0
21)
(0.0
26)
(0.0
02)
(0.0
22)
(0.0
21)
(0.0
02)
yit−
1∗
s it−
1-0
.007
*-0
.080
*0.
002
(0.0
10)
(0.0
32)
(0.0
02)
s it−
10.
347*
0.06
60.
094
-0.0
18(0
.144
)(0
.159
)(0
.187
)(0
.018
)
R2
(%)
20.8
7022
.318
41.9
2515
.787
21.2
7822
.562
44.4
7416
.184
Adj
.R
2(%
)15
.450
16.9
9737
.977
10.0
6215
.715
17.0
8940
.579
10.3
04
Table4:
Testingtheinteraction
between
solven
cyan
dprofitability.
Estim
ates
from
pooled
OLS
regression
withba
nkdu
mmies,
timetrends,an
dheterogeneou
sAR
parameters.
Pan
elA:mod
elof
eq.(5)witho
utstatevariab
le.
Pan
elB:mod
elof
eq.
(5)with
statevariab
les i
t=
1 {S
RIS
Kit
>0}.
Dep
endent
variab
les:
y it
=ln
(STDebt i
t),z i
t=
ln(STAssets i
t),(NI/TA
) it
=NetIncome i
t/TotalAssets i
t,(SRISK/TA
) it
=SRISK
it/TotalAssets i
t.Rob
uststan
dard
errors
inpa
rentheses.
*sign
ificant
parameter
at5%
;**at
1%.Sa
mple:
2107
panelo
bs.over
2000
Q1-20
13Q1(unb
alan
ced),
44ba
nks.
SRISK
istheexpe
cted
capitals
hortfallof
theba
nkin
acrisis.
28
CPP/TA TLGP/TACst 4.280 -7.478
(6.088) (8.948)
T1LVGR 0.222 -0.941*(0.269) (0.395)
T1CR -0.285* 0.181(0.106) (0.155)
log(TA) -0.042 0.806*(0.240) (0.352)
R2 (%) 37.701 72.217Adj. R2 (%) 23.325 65.805
Table 5: Testing factors explaining capital and liquidity injections under theTARP. Estimates from cross-sectional OLS regression. Dependent variables: the amountof capital received under the CPP divided by total assets (CPP/TA), the amount of totalunsecured debt guaranteed by the FDIC divided by total assets (TLGP/TA). T1LVGR: Tier1 Leverage ratio, T1CR: Tier 1 Common Capital Ratio, log(TA): logarithm of total assets,as of 2008Q2. Robust standard errors in parentheses. * significant parameter at 5%; ** at1%. Sample: 17 banks.
29
Dep. variable: yit zit (NI/TA)it (SRISK/TA)itFedfund ratet−1 0.045** 0.001 -0.031 -0.005
(0.011) (0.014) (0.019) (0.003)
Treasury slopet−1 0.077** 0.013 -0.055 0.006**(0.023) (0.026) (0.029) (0.001)
TEDt−1 0.003 0.043 -0.172** 0.009*(0.015) (0.024) (0.048) (0.004)
VIXt−1 0.003** 0.0004 -0.002 -0.00005(0.001) (0.001) (0.001) (0.0002)
M2Gt−1 -4.308** -0.366 0.255 0.154(1.351) (1.035) (1.078) (0.171)
MTGt−1 3.760** -1.281 0.946 -0.748*(1.120) (1.455) (1.236) (0.368)
MMGt−1 0.463** -0.308 -0.197 0.008(0.172) (0.223) (0.336) (0.034)
MMA1t−1 -1.994** 1.058 -0.592 -0.300**(0.455) (0.601) (0.628) (0.073)
MMA2t−1 0.265 -0.291 -1.510** -0.181*(0.222) (0.383) (0.509) (0.086)
R2 (%) 21.996 23.816 44.559 19.217Adj. R2 (%) 16.399 18.350 40.609 13.461
Table 6: Testing common factors. Estimates from pooled OLS regression with bankdummies, time trends, heterogeneous AR parameters and common factors. Robust standarderrors in parentheses. * significant parameter at 5%; ** at 1%. Sample: 2107 panel obs. over2000Q1-2013Q1 (unbalanced), 44 banks. Treasury slope is the slope factor of the Treasuryyield curve. M2G: money supply growth (M2). MTG: mortgage assets growth. MMG:MMMF assets growth. MMA1: proportion of MMMF assets allocated to time deposits.MMA2: proportion of MMMF assets allocated to Treasury, agency, or municipal bonds.
30
NoCom
mon
Factor
Com
mon
Factors
Com
mon
CorrelatedEffe
cts
Dep
.va
riab
le:
yit
z it
NI/
TA
SRIS
K/T
Ay
itz i
tN
I/TA
SRIS
K/T
Ay
itz i
tN
I/TA
SRIS
K/T
A
(SR
ISK
/TA
) it−
1-0.935**
-0.120
-1.681**
-0.847**
-0.178
-1.417**
-0.900**
-0.137
-1.649**
(0.261)
(0.101)
(0.080)
(0.318)
(0.096)
(0.117)
(0.280)
(0.106)
(0.093)
(SR
ISK
/TA
) it−
1∗
s it−
1-0.408
1.757*
-5.871**
-0.687
1.291
-5.108**
-1.413
2.384*
-5.250**
(0.751)
(0.767)
(1.767)
(0.853)
(0.861)
(1.736)
(0.974)
(1.009)
(1.976)
(NI/
TA
) it−
19.704**
-7.944*
-2.809**
8.111*
-7.564*
-2.639*
8.512*
-7.964*
-2.205**
(3.290)
(3.716)
(0.935)
(3.677)
(3.575)
(1.026)
(3.542)
(3.905)
(0.771)
(NI/
TA
) it−
1∗
s it−
1-9.902*
6.315
2.134*
-8.087
4.817
2.034*
-7.181
7.926
2.105*
(4.396)
(5.183)
(0.937)
(4.788)
(4.920)
(0.976)
(4.655)
(5.512)
(0.826)
zit−
1-0.033
-0.035
-0.001
-0.008
-0.047*
-0.004
0.0004
-0.045
-0.004
(0.022)
(0.022)
(0.003)
(0.024)
(0.024)
(0.003)
(0.024)
(0.024)
(0.003)
zit−
1∗
s it−
1-0.021*
0.078*
-0.0008
-0.018*
0.052*
0.0003
-0.013
0.047
0.0006
(0.008)
(0.032)
(0.002)
(0.008)
(0.025)
(0.002)
(0.008)
(0.025)
(0.001)
yit−
1-0.002
-0.031
0.008**
0.002
0.015
0.009**
0.004
0.026
0.008**
(0.022)
(0.021)
(0.002)
(0.002)
(0.018)
(0.003)
(0.020)
(0.018)
(0.003)
yit−
1∗
s it−
1-0.007*
-0.080*
0.002
-0.010
-0.051*
-0.001
-0.013
-0.046*
-0.003
(0.010)
(0.032)
(0.002)
(0.009)
(0.022)
(0.002)
(0.010)
(0.022)
(0.002)
s it−
10.347*
0.066
0.094
-0.018
0.327*
0.098
0.034
0.011
0.253
0.148
0.014
0.029
(0.144)
(0.159)
(0.187)
(0.018)
(0.140)
(0.154)
(0.139)
(0.016)
(0.140)
(0.158)
(0.145)
(0.017)
R2(%
)21.278
22.562
44.474
16.184
25.079
24.247
48.438
20.665
26.730
27.265
50.295
30.990
Adj.
R2(%
)15.715
17.089
40.579
10.304
19.416
18.521
44.567
14.709
21.192
21.767
46.564
25.809
Table7:
Rob
ustnessof
thesolven
cy-liquiditynexusto
common
factors.
Estim
ates
from
pooled
OLS
regres-
sion
with
bank
dummies,
timetrends,heterogeneou
sAR
parametersan
dstatevariab
les i
t=
1 {S
RIS
Kit
>0}.
Dep
en-
dent
variab
les:y i
t=
ln(STDebt i
t),z i
t=
ln(STAssets i
t),
(NI/TA
) it
=NetIncome i
t/TotalAssets i
t,(SRISK/TA
) it
=SRISK
it/TotalAssets i
t.NoCom
mon
Factor:regression
witho
utcommon
factors(eq.
(5)).Com
mon
Factors:
regression
withall(
lagg
ed)common
factorsof
Table6(eq.
(8)).Com
mon
CorrelatedEffe
cts:
regression
withcommon
correlated
effects
(eq.
(9)).Rob
uststan
dard
errors
inpa
rentheses.
*sign
ificant
parameter
at5%
;**
at1%
.Sa
mple:
2107
panel
obs.
over
2000
Q1-20
13Q1(unb
alan
ced),4
4ba
nks.
SRISK
istheexpe
cted
capitals
hortfallof
theba
nkin
acrisis.
31
Dep. variable: dYit dZit yit zit (NI/TA)it (SRISK/TA)it
(SRISK/TA)it−1 -0.013 -0.039 -1.059** -0.038 -2.241**(0.024) (0.038) (0.235) (0.118) (0.282)
(NI/TA)it−1 -0.870 0.290 2.313 -4.185 -1.211**(0.592) (0.508) (2.232) (2.348) (0.386)
zit−1 -0.035** -0.006 -0.034 0.162 -0.002(0.005) (0.006) (0.024) (0.162) (0.002)
yit−1 -0.0004 -0.018** -0.002 -0.023 0.008**(0.006) (0.006) (0.022) (0.021) (0.002)
dZit−1 0.101* 0.004 -0.110 -0.103 0.014(0.046) (0.109) (0.116) (0.145) (0.012)
dY it−1 -0.104* -0.137 0.157 0.254 -0.002(0.052) (0.199) (0.096) (0.224) (0.011)
R2 (%) 11.319 12.008 21.047 22.411 42.099 15.837Adj. R2 (%) 5.197 5.934 15.554 17.013 38.099 10.024
Table 8: Estimates from pooled OLS regression with bank dummies, time trends and het-erogeneous AR parameters. Dependent variables: dYit = ln(LTDebtit/LTDebtit−1), dZit =ln(LTAssetsit/LTAssetsit−1), yit = ln(STDebtit), zit = ln(STAssetsit), (NI/TA)it =NetIncomeit/TotalAssetsit, (SRISK/TA)it = SRISKit/TotalAssetsit. Robust standarderrors in parentheses. * significant parameter at 5%; ** at 1%. Sample: 2107 panel obs.over 2000Q1-2013Q1 (unbalanced), 44 banks. SRISK is the expected capital shortfall of thebank in a crisis.
32
PA
NEL
A:Fo
reca
stin
gth
esh
ort-
term
deb
tTrend
assumption
AR
VAR
INT
CF
4RMSF
E(SRISK)4RMSF
E(N
I)4RMSF
E(STA)
heterogeneou
strends
1856
716
338
1626
711
939
1871
-236
451
homogenou
strend
1023
910
577
7943
1075
381
3-5
19-1
98no
trend
8438
7943
6801
9046
1165
-185
-78
trendbreak
1003
882
9575
8112
986
1252
-43
-148
PA
NEL
B:Fo
reca
stin
gth
esh
ort-
term
asse
tsTrend
assumption
AR
VAR
INT
CF
4RMSF
E(SRISK)4RMSF
E(N
I)4RMSF
E(STA)
heterogeneou
strends
1307
913
817
1359
615
011
-113
84-8
02ho
mogenou
strend
1347
713
828
1357
014
381
45-1
27-2
28no
trend
1463
215
349
1448
314
176
370
-192
-651
trendbreak
1323
513
108
1298
813
985
-60
-101
264
PA
NEL
C:Fo
reca
stin
gth
eliqu
idas
set
shor
tfal
l(w
hol
esa
mple
)Trend
assumption
AR
VAR
INT
CF
4RMSF
E(SRISK)4RMSF
E(N
I)heterogeneou
strends
1983
417
374
1567
313
426
1638
-301
homogenou
strend
1573
718
475
1689
115
244
370
-791
notrend
1599
917
172
1595
214
915
1629
-576
trendbreak
1435
314
159
1399
315
782
225
-321
PA
NEL
D:Fo
reca
stin
gth
eliqu
idas
set
shor
tfal
lof
adeq
uat
ely
capit
aliz
edban
ks(S
RIS
Kit≤
0)Trend
assumption
AR
VAR
INT
CF
4RMSF
E(SRISK)4RMSF
E(N
I)heterogeneou
strends
7702
6443
6992
4927
87-2
7ho
mogenou
strend
5294
6003
6672
5338
-185
-84
notrend
5297
6117
6666
5725
-58
-54
trendbreak
4431
5374
5654
5081
-236
-49
PA
NEL
E:Fo
reca
stin
gth
eliqu
idas
set
shor
tfal
lof
capit
al-c
onst
rain
edban
ks(S
RIS
Kit
>0)
Trend
assumption
AR
VAR
INT
CF
4RMSF
E(SRISK)4RMSF
E(N
I)heterogeneou
strends
2541
322
329
1985
817
267
2214
-406
homogenou
strend
2033
723
916
2161
519
658
536
-105
7no
trend
2069
222
122
2032
619
125
2226
-774
trendbreak
1862
418
170
1787
720
439
358
-429
Table9:
Forecastingtheshort-term
balan
cesheet.
Roo
tMean
Squa
reFo
recastingError
(RMSF
E):
one-step
aheadforecastingover
2011
.Fixed
estimationsample(200
0-20
10),
inform
ationup
datedeach
quarter.
AR:un
ivariate
autoregressive
mod
el.VA
R:pa
nelVA
Rmod
el(eq.
(4)).IN
T:pa
nelVA
RwithinteractionwithSR
ISK
(eq.
(5)).CF:
panelV
ARwithinteractionwithSR
ISK
andcommon
factors(eq.
(8)).4RMSE
(x)istheincrease
inRMSE
whenvariab
lexis
notinclud
edin
theVA
Rmod
el.Liqu
idassetshortfall=
STDebt-ST
Assets.
Inbo
ld:minim
umRMSF
Eforeach
line(trend
assumption).
33
STAssets STDebt STAssetsSTDebt
LTAssets LTDebtLTAssets LTDebt
EquityEquity
Bank is balance-sheet insolvent.Assets are liquidated at a loss, only senior creditors are repaid.
STDebt
LTAssetsLTDebt
Figure 1: Simplified balance sheet. Liquid asset shortfallit = STDebtit−STAssetsit. Capitalshortfallit = k ∗ (STAssetsit +LTAssetsit)−Equityit. Expected capital shortfall in a crisisSRISKit = E [k ∗ (STAssetsit+h + LTAssetsit+h)− Equityit+h|crisist+h] = k∗(LTDebtit +STDebtit) − (1 − k) ∗ Equityit ∗ (1 + E(Rit+h|crisist+h)), where k is the prudential capitalratio (8%).
34
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
0
25000
50000
75000
100000
125000
Av
erag
e li
qu
id a
sset
sh
ort
fall
($
m)
Avg(STDebt STAssets) if SRISK<0 Avg(STDebt STAssets) if SRISK>0
#firms (SRISK<0) #firms (SRISK>0)
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
20
40#firms (SRISK<0) #firms (SRISK>0)
Figure 2: Cross-sectional average of the liquid asset shortfall of capital-constrained banks(black line) vs. cross-sectional average of the liquid asset shortfall of adequately capitalizedbanks (dashed line). Liquid asset shortfall = Short-term debt - Short-term assets ($m).“Adequately capitalized” means SRISKit ≤ 0.
35
Avg(STDebt) Avg(STAssets)
2000 1 2 3 4 5 6 7 8 9 10 11 12 13
20000
40000
60000
Avg(STDebt) Avg(STAssets)
Avg(STDebt STAssets)
2000 1 2 3 4 5 6 7 8 9 10 11 12 13
10000
0
10000
20000
30000
40000 Avg(STDebt STAssets)
Avg(LTDebt) Avg(LTAssets) Avg(TotalAssets)
2000 1 2 3 4 5 6 7 8 9 10 11 12 13
100000
150000
200000
250000
300000 Avg(LTDebt) Avg(LTAssets) Avg(TotalAssets)
Avg(NetIncome)
2000 1 2 3 4 5 6 7 8 9 10 11 12 13
500
0
500
Avg(NetIncome)
Figure 3: Cross-sectional averages of the balance sheet (in $m): short-term debt andshort-term assets (top-left panel), difference between short-term debt and short-term assets(top-right panel), total assets and long-term balance sheet (bottom-left panel), net income(bottom-right panel).
36
Avg(T1CR)
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
0.10
0.12
0.14Avg(T1CR)
Avg(T1LVGR)
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
0.08
0.09
0.10Avg(T1LVGR)
Avg(SRISK) ($m) Avg(max(0,SRISK)) ($m)
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
0
10000Avg(SRISK) ($m) Avg(max(0,SRISK)) ($m)
Figure 4: Cross-sectional averages of solvency risk measures. T1CR is the Tier 1 commoncapital ratio (Tier 1 common capital divided by risk-weighted assets); T1LVGR is the Tier 1leverage ratio (Tier 1 capital divided by total assets); SRISK is the expected capital shortfallin a crisis.
37
0 5 10
1
3NI > NI
0 5 10
2
1
0
1SRISK > NI
0 5 10
1.0
0.5
0.0STD > NI
0 5 10
0.1
0.0
0.1
STA > NI
0 5 10
0.4
0.2
0.0
NI > SRISK
0 5 10
0.0
0.5
1.0
SRISK > SRISK
0 5 10
0.1
0.2
0.3STD > SRISK
0 5 10
0.03
0.01
STA > SRISK
0 5 10
0.5
1.0
1.5
2.0NI > STD
0 5 10
4
2
0SRISK > STD
0 5 10
0
5
10STD > STD
0 5 10
0.50
0.25
0.00STA > STD
0 5 10
1.0
0.5
0.0NI > STA
0 5 10
0.1
0.2
0.3SRISK > STA
0 5 10
0.0
0.1
0.2
0.3STD > STA
0 5 10
2.5
5.0
7.5
STA > STA
Figure 5: Impulse response functions. Median impulse response function (black lines) be-tween the 25% and 75% impulse response quantiles (dotted lines).
38
0 5 10
0
1
2
NI > NI
0 5 10
5
0SRISK > NI
0 5 10
1.0
0.5
0.0STD > NI
0 5 10
0.0
0.5
STA > NI
0 5 10
0.50
0.25
0.00
NI > SRISK
0 5 10
0.25
0.75
SRISK > SRISK
0 5 10
0.10
0.15
STD > SRISK
0 5 10
0.03
0.01STA > SRISK
0 5 10
1
2
NI > STD
0 5 10
4
2
SRISK > STD
0 5 10
2.5
5.0
7.5
STD > STD
0 5 10
0.6
0.4
0.2
STA > STD
0 5 10
1.0
0.5
0.0NI > STA
0 5 10
0.50
0.25
0.00
0.25SRISK > STA
0 5 10
0.0
0.2STD > STA
0 5 10
2.5
5.0
7.5STA > STA
Figure 6: Impulse response functions with SRISK as a state variable (eq. (5)). Medianimpulse response function when SRISKit ≤ 0 (black line) and median impulse responsefunction when SRISKit > 0 (dashed line).
39
Actual Forecast
2005 2006 2007 2008 2009 2010 2011 2012 2013
30000
40000
50000
60000
70000
Avg(STDebt) ($m)
Actual Forecast
2005 2006 2007 2008 2009 2010 2011 2012 2013
20000
30000
40000
50000
60000
Avg(STAssets) ($m)
2005 2006 2007 2008 2009 2010 2011 2012 2013
5000
0
5000
Avg(STDebt flows) ($m)
2005 2006 2007 2008 2009 2010 2011 2012 2013
5000
0
5000
Avg(STAssets flows) ($m)
Figure 7: Forecasting the short-term balance sheet over 2011Q1-2013Q1: dynamic forecasts(panel VAR with SRISK as a state variable (eq. (5)), break in trend).
40
Actual Forecast
2005 2006 2007 2008 2009 2010 2011 2012 2013
10000
0
10000
20000
30000
40000Avg(STDebt STAssets) ($m)
Actual Forecast
SRISK<0 SRISK>0
2005 2006 2007 2008 2009 2010 2011 2012 2013
0
25000
50000
75000
100000
125000
Avg(STDebt STAssets) ($m)
SRISK<0 SRISK>0
Figure 8: Forecasting the Liquid Asset Shortfall over 2011Q1-2013Q1: dynamic forecasts(panel VAR with SRISK as a state variable (eq. (5)), break in trend).
41
Appendix
A Short-term debt and short-term assets compositionComposition of the short term debt estimate (SNL definitions):
• Fed funds purchased: The gross dollar amount of funds borrowed in the form of imme-diately available funds under agreements or contracts that mature in one business dayor roll over under a continuing contract, regardless of the nature of the transaction orthe collateral involved. Includes securities sold under agreements to repurchase thatinvolve the receipt of immediately available funds and mature in one business day orroll over under a continuing contract.
• Repurchase agreements: The gross dollar amount of security repurchase agreementsthat mature in more than one business day, other than securities sold under repurchaseagreements to maturity, but including sales of participations in pools of securities thatmature in more than one business day.
• Brokered Deposits (< $100K, maturity ≤ 1 Year): Brokered deposits issued in de-nominations of less than $100,000 with a remaining maturity of one year or less andare held in domestic offices of commercial banks or other depository institutions thatare subsidiaries of the reporting bank holding company. Remaining maturity is theamount of time remaining from the report date until the final contractual maturity ofa brokered deposit.
• Time Deposits (≥ $100K, maturity≤ 1Year): Time deposits issued in denominations of$100,000 or more with a remaining maturity of one year or less. Remaining maturity isthe amount of time remaining from the report date until the final contractual maturityof a time deposit.
• Foreign Office Time Deposits (maturity ≤ 1Year): All time deposits in foreign officeswith remaining maturities of one year or less. Remaining maturity is the amount oftime remaining from the report date until the final contractual maturity of a timedeposit.
• Commercial Paper: The total amount outstanding of commercial paper issued by thereporting bank holding company or its subsidiaries.
• Other borrowed money: The total amount of money borrowed by the consolidatedbank holding company with a remaining maturity of one year or less. For purposes ofthis item, remaining maturity is the amount of time remaining from the report dateuntil the final contractual maturity of a borrowing without regard to the borrowing’srepayment schedule, if any. Includes the dollar amount outstanding of all interest-bearing demand notes issued to the U.S. Treasury by the depository institutions thatare consolidated subsidiaries of the reporting bank holding company. Also includes
42
mortgage indebtedness and obligations under capitalized leases with a remaining ma-turity of one year or less. Also includes the total amount of money borrowed with aremaining maturity of one year or less: (1) on its promissory notes; (2) on notes andbills rediscounted; (3) on loans sold under repurchase agreements that mature in morethan one business day; (4) by the creation of due bills representing the bank holdingcompany’s receipt of payment and similar instruments, whether collateralized or un-collateralized; (5) from Federal Reserve Banks; (6) by overdrawing ’due from’ balanceswith depository institutions, except overdrafts arising in connection with checks ordrafts drawn by subsidiary depository institutions of the reporting bank holding com-pany and drawn on, or payable at or through, another depository institution either ona zero-balance account or on an account that is not routinely maintained with suffi-cient balances to cover checks or drafts drawn in the normal course of business duringthe period until the amount of the checks or drafts is remitted to the other depositoryinstitution; (7) on purchases of so-called ’term federal funds’; and (8) on any otherobligation for the purpose of borrowing money that has a remaining maturity of oneyear or less and that is not reported elsewhere.
0
500000
1000000
1500000
2000000
2500000
3000000
3500000
2000
Q1
2000
Q3
2001
Q1
2001
Q3
2002
Q1
2002
Q3
2003
Q1
2003
Q3
2004
Q1
2004
Q3
2005
Q1
2005
Q3
2006
Q1
2006
Q3
2007
Q1
2007
Q3
200
8Q
1
2008
Q3
2009
Q1
2009
Q3
2010
Q1
2010
Q3
2011
Q1
2011
Q3
2012
Q1
2012
Q3
2013
Q1
14. Total fed funds & repurchase agreements 1. Brokered deposits < $100K with <= 1yr maturity
3. Time deposits >= $100K with <= 1yr maturity 4. Foreign office time deposits <= 1yr maturity
a. Commercial paper b. Other with maturity of < 1 year
Figure 9: Short term debt composition ($m) - 44 BHCs
43
Composition of the short term assets estimate (SNL definitions):
• Cash & Non interest-bearing Deposits: The total of all noninterest-bearing balancesdue from depository institutions, currency and coin, cash items in process of collection,and unposted debits. Includes balances due from banks in the U.S., banks in foreigncountries and foreign central banks, foreign branches of other U.S. banks, Federal HomeLoan Banks, and Federal Reserve Banks.
• Total Interest-bearing Balances: The total of all interest-bearing balances due fromdepository institutions and foreign central banks that are held in offices of the bankholding company or its consolidated subsidiaries.
• Fed Funds Sold: The gross dollar amount of funds lent in the form of immediatelyavailable funds under agreements or contracts that mature in one business day or rollover under a continuing contract. Includes securities purchased under agreements toresell that involve the receipt of immediately available funds and mature in one businessday or roll over under a continuing contract.
• Reverse Repurchases Agreements: The gross dollar amount of security resale agree-ments that mature in more than one business day, other than securities purchasedunder resale agreements to maturity, and of purchases of participations in pools ofsecurities that mature in more than one business day.
• Debt Securities Maturing or Repriced (maturity ≤ 1Year): All securities held by theconsolidated bank holding company with a remaining maturity or amount of timeremaining until next repricing date of one year or less. Held-to-maturity securitiesare reported at amortized cost and available-for-sale securities are reported at fairvalue. Remaining maturity is the amount of time remaining from the report date untilthe final contractual maturity of the instrument without regard to the instrument’srepayment schedule. Next repricing date is the date the interest rate on a floating ratedebt security can next change. (Y9 Line Item: BHCK0383)
44
0
500000
1000000
1500000
2000000
2500000
2000
Q1
2000
Q3
2001
Q1
2001
Q3
2002
Q1
2002
Q3
2003
Q1
2003
Q3
2004
Q1
2004
Q3
2005
Q1
2005
Q3
2006
Q1
2006
Q3
2007
Q1
2007
Q3
200
8Q
1
2008
Q3
2009
Q1
2009
Q3
2010
Q1
2010
Q3
2011
Q1
2011
Q3
2012
Q1
2012
Q3
2013
Q1
2. a. Debt securities: Maturity/Repricing <= 1Yr a. Cash & noninterest bearing balances
b. Total interest bearing balances 3. Tot fed funds & reverse repos
Figure 10: Short-term assets composition ($m) - 44 BHCs
45
B Stationarity of balance sheet aggregatesTo test for the stationarity of yit, zit and other balance sheet quantities, I apply the unit roottest of Pesaran (2007) (CIPS) robust to cross-sectional dependence between individuals inthe panel data set. The the null hypothesis is H0 : α21 = α22 = ... = α2N = 0, i = 1, 2, ..., N(unit root), and the alternative Ha : α21 < 0, ..., α2N0 < 0, N0 ≤ N(a significant fraction ofthe panel is stationary). The regression for the CIPS unit root test is
dyit = α0i + α1idyit−1 + α2iyit−1 + aidyt + biyt−1 + cidyt−1 + θit+ εit, (10)
where dyt = N−1∑N
i=1 dyit, yt = N−1∑N
i=1 yit. The CIPS test statistics are reported in Table10 for both cases with and without trend (i.e. θi = 0, ∀i). Based on the CIPS statistics andgiven the critical values of the CADF distribution, yit is stationary only when the regressionincludes a trend. The hypothesis of the absence of a trend is rejected based on a Wald test,therefore yit is considered stationary in the rest of the paper.
On the other bank sheet aggregates, the UR hypothesis is not rejected for the size (loga-rithm of total assets) and the long-term balance sheet (logarithm of long-term assets Zit andlong-term debt Yit). Finally, the short-term assets, SRISK and the net income divided bytotal assets are stationary with this test.
Intercept only Intercept and trendCIPS CIPSb CIPS CIPSb
yit -2.064 -1.922 -2.725 -2.660zit -2.538 -2.545 -2.798 -2.849
NIit/TAit -3.541 -3.831 -4.101 -4.381Yit -2.071 -2.199 -2.274 -2.468Zit -1.954 -2.098 -2.449 -2.584
log(TAit) -1.709 -1.932 -2.163 -2.336SRISKit/TAit -2.579 -2.434 -2.951 -2.989
Table 10: Panel UR tests: CIPS statistics. CADF 5% critical values: -2.11 (intercept only),-2.60 (intercept and trend). CIPSb is the CIPS statistic based on a balanced panel dataset.yit = ln(STDebtit), zit = ln(STAssetsit), Yit = ln(LTDebtit), Zit = ln(LTAssetsit), NIit:net income, TAit: total assets.
46
C Robustness checks
C.1 Period dummies
Dep. variable: yit zit (NI/TA)it (SRISK/TA)it yit zit (NI/TA)it (SRISK/TA)it
(SRISK/TA)it−1 -1.063** -0.028 -2.225** -0.623 -0.108 -2.033**(0.245) (0.118) (0.301) (0.357) (0.178) (0.217)
(SRISK/TA)it−1 ∗ ct -0.606** 0.188 -0.321(0.231) (0.167) (0.182)
(SRISK/TA)it−1 ∗ pct -0.091 -0.298 0.137(0.383) (0.262) (0.194)
(NI/TA)it−1 2.354 -4.228 -1.217** 25.036** -7.046 -7.529**(2.278) (2.331) (0.389) (7.838) (5.344) (2.011)
(NI/TA)it−1 ∗ ct -24.850** 5.882 6.355**(8.248) (6.315) (1.576)
(NI/TA)it−1 ∗ pct -20.939 -8.164 9.165**(12.332) (7.475) (2.463)
zit−1 -0.038 -0.015 -0.002 -0.012 -0.033 0.001(0.023) (0.020) (0.002) (0.024) (0.024) (0.002)
zit−1 ∗ ct -0.003 0.018 -0.006**(0.011) (0.033) (0.002)
zit−1 ∗ pct -0.008 0.003 -0.001(0.020) (0.030) (0.003)
yit−1 -0.004 -0.067** 0.008** 0.002 0.015 0.008*(0.021) (0.026) (0.002) (0.023) (0.017) (0.003)
yit−1 ∗ ct 0.011 -0.048 0.007*(0.014) (0.034) (0.003)
yit−1 ∗ pct 0.009 -0.059* 0.00001(0.022) (0.029) (0.004)
ct 0.114 -0.151 0.360 -0.017(0.168) (0.241) (0.227) (0.028)
pct 0.023 -0.042 0.904** 0.005(0.296) (0.383) (0.237) (0.034)
R2 (%) 20.870 22.318 41.925 15.787 24.877 22.956 44.361 25.551Adj. R2 (%) 15.450 16.997 37.977 10.062 19.405 17.343 40.336 20.165
Table 11: Estimates from pooled OLS regression with bank dummies, time trends and het-erogeneous AR parameters. Dependent variables: yit = ln(STDebtit), zit = ln(STAssetsit),(NI/TA)it = NetIncomeit/TotalAssetsit, (SRISK/TA)it = SRISKit/TotalAssetsit. Ro-bust standard errors in parentheses. * significant parameter at 5%; ** at 1%. Sample: 2107panel obs. over 2000Q1-2013Q1 (unbalanced), 44 banks. SRISK is the expected capitalshortfall of the bank in a crisis.
47
C.2 Short-term debt components
Dep. variable: yit FFRepo Br Dep Time Dep For Dep ComPaper OtherBor(SRISK/TA)it−1 -1.063** -1.217** 0.147 -0.363* -1.155* -2.330** -0.281
(0.245) (0.403) (0.518) (0.175) (0.531) (0.364) (0.377)
(NI/TA)it−1 2.354 0.152 -11.694 -1.660 10.880 17.540 3.249(2.278) (2.437) (8.376) (4.733) (6.282) (10.392) (9.375)
zit−1 -0.038 0.015 -0.028 -0.046* -0.191* -0.091 -0.233**(0.023) (0.051) (0.096) (0.019) (0.093) (0.083) (0.079)
# obs. 2107 1979 950 2096 1337 966 2035# banks 44 44 40 44 34 27 44R2 (%) 20.870 19.723 23.649 34.947 38.600 25.330 22.656
Adj. R2 (%) 15.450 13.843 12.279 30.459 33.355 18.187 17.152
Table 12: Estimates from pooled OLS regression with bank dummies, time trends and het-erogeneous AR parameters. Dependent variables: log of the different components of theshort term debt (see definitions in Appendix A): Fed funds and Repos (FFRepo), BrokeredDeposits (Br Dep), uninsured Time Deposits (Time Dep), Foreign Deposits (For Dep), Com-mercial Papers (ComPaper) and Other Borrowed Money (OtherBor). Robust standard errorsin parentheses. Robust standard errors in parentheses. * significant parameter at 5%; ** at1%. Sample: 2107 panel obs. over 2000Q1-2013Q1 (unbalanced), 44 banks. SRISK is theexpected capital shortfall of the bank in a crisis.
48
D Sample of banks
Name Ticker SNL ID RSSD ID Industry Market CapAmerican Express Company AXP 102700 1275216 Specialty Lender 60,834Bank of America Corporation BAC 100369 1073757 Bank 183,125The Bank of New York Mellon Corporation BK 100144 3587146 Bank 55,522BB&T Corporation BBT 100438 1074156 Bank 16,852Capital One Financial Corp. COF 103239 2277860 Bank 18,215Citigroup, Inc. C 4041896 1951350 Bank 146,644Fifth Third Bancorp FITB 100260 1070345 Bank 13,386The Goldman Sachs Group, Inc. GS 4039450 2380443 Broker Dealer 85,520JPMorgan Chase & Co. JPM 100201 1039502 Bank 146,622KeyCorp KEY 100334 1068025 Bank 9,117MetLife, Inc. MET 4051708 2945824 Insurance 45,636Morgan Stanley MS 103042 2162966 Broker Dealer 56,362The PNC Financial Services Group, Inc. PNC 100406 1069778 Bank 22,355Regions Financial Corporation RF 100233 3242838 Bank 16,439State Street Corporation STT 100447 1111435 Bank 31,360SunTrust Banks, Inc. STI 100449 1131787 Bank 21,756U.S. Bancorp USB 4047176 1119794 Bank 54,804Wells Fargo & Company WFC 100382 1120754 Bank 101,269Franklin Resources Inc. BEN 102719 1246216 Asset Manager 28,037Commerce Bancshares, Inc. CBSH 100184 2815235 Bank 3,229CIT Group Inc. CIT 102820 1036967 Specialty Lender NAComerica Incorporated CMA 100206 1029259 Bank 6,574Huntington Bancshares Incorporated HBAN 100307 1068191 Bank 5,401Marshall & Ilsley MI 100364 3594612 Bank 7,086M&T Bank Corporation MTB 100253 1037003 Bank 8,708National City Corp. NCC 100378 1069125 Bank 10,433Northern Trust Corporation NTRS 100386 1199611 Bank 16,843New York Community Bancorp Inc. NYCB 1024119 2132932 Savings/Thrift/Mutual 5,689The Charles Schwab Corporation SCHW 102775 1026632 Broker Dealer 29,547Synovus Financial Corporation SNV 100440 1078846 Bank 7,943UnionBanCal Corporation UB 1022285 1378434 Bank 6,776Wachovia Bank WB 100293 1073551 Bank 75,122Zions Bancorp. ZION 100501 1027004 Bank 4,995Associated Banc-Corp ASBC 100135 1199563 Bank 3,442Bank of Hawaii Corporation BOH 100161 1025309 Bank 2,506BOK Financial Corporation BOKF 100003 1883693 Bank 3,471Popular, Inc. BPOP 100165 2138466 Bank 2,971Cullen/Frost Bankers, Inc. CFR 100196 1102367 Bank 2,963
Table 13: Sample 1/2. Market capitalization in $m (Dec 30, 2007).
49
Name Ticker SNL ID RSSD ID Industry Market CapCity National Corporation CYN 100225 1131004 Bank 2,866Discover Financial Services DFS 4096334 3846375 Specialty Lender NAEast West Bancorp, Inc. EWBC 4040606 2734233 Bank 1,527First Citizens BancShares, Inc. FCNCA 100247 1105470 Bank 1,619First Horizon National Corporation FHN 100292 1094640 Bank 2,294Fulton Financial Corporation FULT 100294 1117129 Bank 1,946Hancock Holding Company HBHC 100308 1086533 Bank 1,207Prosperity Bancshares, Inc. PB 1018962 1109599 Bank 1,297SVB Financial Group SIVB 100433 1031449 Bank 1,673TCF Financial Corporation TCB 102002 2389941 Bank 2,272Webster Financial Corporation WBS 102030 1145476 Bank 1,710
Table 14: Sample 2/2. Market capitalization in $m (Dec 30, 2007).
50